Share this page!

Enter URL or ID to Unroll

You can paste full URL like: https://x.com/threadreaderapp/status/1644127596119195649
or just the ID like: 1644127596119195649

How to get URL link on X (Twitter) App

  1. On the Twitter thread, click on or icon on the bottom
  2. Click again on or Share Via icon
  3. Click on Copy Link to Tweet
  4. Paste it above and click "Unroll Thread"!
  5. More info at Twitter Help
Michael Cohen Profile picture
Michael Cohen
@mc_anthropic
Working on the @AnthropicAI API. Views are my own
Apr 8 • 4 tweets • 3 min read
Hi, I'm Michael, an engineer working on Anthropic's API team. I'm really excited to finally be able to talk about Claude Managed Agents, a suite of APIs and developer tooling we've been working on for the past few months. Image At Anthropic we've been working on building agents for a while now. Like everyone else in the industry we realized that building and deploying remote, long running agents at scale is NOT an easy undertaking.

The Messages API has been the gateway for folks to get access Claude programmatically, but it's a stateless inference API focused on "messages in -> messages out".
Coding agents taught us that LLMs can do meaningfully useful work when given access to a computer, but it's not easy to then convert that to a multi-tenant fleet of Claudes running for long periods of time.

Running entire fleets of agents, each with their own security, state, tools, skills, sandboxing infrastructure, MCP servers, authentication credentials, and more takes a lot of time and effort to get right. Building that infrastructure on your own slows you down from focusing on your own core product experience, and requires an investment in resources that many teams just don't have. What's worse is that the moment some new agent paradigm is unlocked, you now have to go back and invest more time in reworking your harness to best utilize the latest and greatest.